Start
Quickstart
From nothing to a gateway in front of your agent in about five minutes.
#1. Get the gateway
Install it with the installer, or use the container. Both are described in Installation.
curl -fsSL <download-address>/install.sh | shrakuvo versionPrefer not to run anything? The hosted gateway needs only a base URL change.
#2. Start the gateway
The gateway holds your real provider key, so your agent never needs it.
export OPENAI_API_KEY=sk-..../rakuvo serve -listen 127.0.0.1:8081 -tool-policy-auto -loop-guard -trace ./traces.jsonl-tool-policy-auto adds the tool-use instruction only for models where it was measured to help. -loop-guard watches for repeating agents. -trace writes one line per model call.
#3. Point your agent at it
Change the base URL. Nothing else.
from openai import OpenAIclient = OpenAI(base_url="http://127.0.0.1:8081/v1", api_key="anything")client.chat.completions.create(model="gpt-5-mini", messages=[...], tools=[...])#4. See what it did
Open traces.jsonl. Each line records the model, tokens, cache hits, latency, and a note for every layer that acted, such as prompt: added the tool-use policy or guard: removed markdown fence.
#5. Add more
./rakuvo serve -listen 127.0.0.1:8081 \ -tool-policy-auto \ -verify-model gpt-4o-mini -verify-mode veto \ -loop-guard \ -context-keep-observations 6 -context-mask-step 3 \ -memory-dir ./memory -memory-model gpt-4o-mini \ -trace ./traces.jsonlEvery flag is listed in the configuration reference.